Generate a video from a text prompt and optionally an image. This is an asynchronous operation that returns a request_id for polling.

Image input for generation and editing requests. Accepts a public URL, a base64-encoded data URL, or a file_id from the xAI Files API.

Optional mid-video keyframe anchors, strictly between the endpoint pins (image as the first frame, last_frame as the last). Each entry pins an image to appear literally at its timestamp. Only supported by select video models; at most 4 entries.

Output destination for generated video.

Optional reference audio clips (voice identity) for reference-to-video generation. Each entry supplies a clip via url or selects a voice via voice_id; caller-supplied clips must be at most 15 seconds long. Only supported by select video models; at most 3 entries. May be provided without reference_images (audio-only reference-to-video) — at least one reference of either kind selects the reference-to-video mode.

Optional reference images for reference-to-video (R2V) generation. When provided generates video using these images as style/content references.

Configuration for storing generation output in the Files API. When provided in a generation request, the output is stored as a permanent file and a file_output reference is included in the response.